Pre-test

35

36 On the other hand, when asked about which criteria would boost a negative effect, there was a significant difference detected on the perception about if this type of Social Media Content strategy can negatively impact annoyance – those exposed to Opportunistic and Brand-To-Brand RTM tended to say “maybe” or “yes” when asked if RTM could become annoying, while for the other types of RTM the respondents said “maybe” or “no”, which matches Kerns (2014) assumption regarding the negative effect that brand-on-brand RTM might have. However, no statistical significance was found on which criteria might contribute to this negative effect. The factor that most contributes to annoyance is the fact that the respondents’ social media platform are full of this type of content – which might underline even more the importance of knowing when it is relevant to use it or not, since it can create the opposite effect.

After the experiment the respondents were asked to mention a brand they know used this type of marketing content – as presented in Table 10, 53% mentioned Control – and when asked if they usually interact with the brand the average score was “maybe” as well as regarding a sense of connection with the brand. However, when asked if preferably they purchase products from that brand the average scoring was two, which means “yes”.

5.3. BCE

CBE C

As previously stated, CBE construct is composed by cognitive, affection and activation (Hollebeek et al., 2014). Also, BCE has been conceptualized as a factor of seven different constructs - Self-identity, Social bonding, Utilitarian, Humor, Aesthetic, Awe-inspiring and Discerning (Waqas et al., 2021). As already defended, Utilitarian was not considered since it does not apply to the context of this study, meaning only six constructs were accessed for this research. In practical terms, this means we are dealing with second order constructs – BCE and CBE – which need previous treatment before being considered for any further analysis.

First checking some correlation issues and possible limiting the amount of variables, factor analysis was run. The extraction method used was principle components analysis based on eigenvalues criteria and the rotation principle was Varimax procedure - orthogonal method of factor rotation that minimizes the number of variables with high loadings on a factor, thereby enhancing the interpretability of the factors (Hair et al., 2014). The results from this procedure – which can be accessed in Table 20 - transformed BCE into four factors as it is possible to observe in Table 4.

37

Constructs Initial

Items Factors Items SSL %

Variance

SSL Cumulative

%

Cronbach’s alpha

BCE 24 BCE_ID.Bound 7 24,46 24,46 0.909

BCE_Aesth.Insp 6 22,02 46,48

BSE_Humor 6 20,63 67,11

BCE_Discerning 5 18,41 85,52

CBE 10 CBE 10 82,21 82,21 0.975

BUI 4 BUI 4 92,05 92,05 0.961

SBC 7 SBC 7 89.80 89.80 0.985

Note: CBE: Consumer-Brand Engagement, BUI: Brand Usage Intent, SBC: Self-Brand Connection, n=32.

Table 5 – Pre-test: Factor Analysis Results

CBE was merged into one single factor, contrary to what was concluded that it would be a second order construct with three factors. BUI and SBC are both first order constructs which didn’t seem surprising to be joined into one single factor.

Given that the sample size is rather low, according to Hair et al. (2014), the factor loadings of each component/factor should be >0.75 to ensure convergent validity. For this reason, 10 items from the BCE construct were eliminated to access differences. However, this extraction resulted on a lower SSL cumulative value (85.52% vs 78.48%) and showed no significant improvement on the AVE and Composite Reliability results within the factors, meaning they should still be considered (Hair, et al., 2014), and, therefore, not removed. The four selected components then explain 85.52% of the variance in BCE, for CBE a single component explains 82.2% of the variance, for BUI 92.05% and SBC 89.8% which are consistent and relevant results.

We examined the factor’s reliability with the help of Cronbach's alpha and composite reliability (CR) value. Cronbach's alpha value of 0.7 or above is recommended (Hair et al., 2014). Furthermore, CR values of 0.70–1 indicate satisfactory to good reliability (Sarstedt et al., 2014). Table 5 shows that all CR and Cronbach's alphas are well above the threshold level.

Constructs α CR AVE BCE_ID.Bound BCE_Aesth.Insp BSE_humor BCE_Discerning CBE BUI SBC BCE_ID.Bound 0,905 0,9047 0,5813 0,754

BCE_Aesth.Insp 0,890 0,8998 0,6058 0.624 0,672

BSE_Humor 0,833 0,8330 0,5568 0.778 0.802 0,768

BCE_Discerning 0,865 0,8646 0,5621 0.578 0.715 0.799 0,748

CBE 0.975 0,9787 0,8220 0.696 0.904 0.894 0.748 0,9046

BUI 0.961 0,9723 0,8979 0.588 0.807 0.766 0.778 0.872 0,9473

SBC 0.985 0,9878 0,9205 0.631 0.914 0.820 0.725 0.932 0.881 0,9593

38

Note: The diagonal (in bold) represents the AVEs of each construct; below the diagonal are the Pearson correlations among constructs; CBE: Consumer-Brand Engagement, BUI: Brand Usage Intent, SBC: Self-Brand Connection, α:Cronbach’s alpha, CR: Construct Reliability, AVE: Average Variance Extracted n = 32.

Table 6 – Pre-test: Construct reliability and validity

After examining the scale's internal consistency, we assessed the convergent validity using Average Variance Extracted (AVE) statistics. An acceptable value of AVE is 0.50 or above (Hair et al., 2014) i.e., all measures “positively correlate with alternative measures of the same construct” (Hair, et al., 2014, p.137). Table 6 indicates that all AVEs values are above 0.5, thus establishing convergent validity. The discriminant validity was also examined by comparing the square root of the AVE values of each construct with the bivariate correlations among all constructs (Fornell & Larcker, 1981), which should be greater than the inter-construct correlations (Hair et al., 2013). Table 6 shows that the square root of all AVEs is greater than the inter-construct correlations, thus establishing the discriminant validity, except for two factors of BCE – which might be due to the non-variable reduction completion process according with the factor loadings criteria suggested by Hair et al. (2014). Again, in this case, to have a deeper understanding of the data while there is still not enough volume of responses, it will be considered.

5.4. M

A

T

I

E

RTM

To access and validate the hypothesis, a multivariate analysis of variance was performed to investigate RTM’s (independent variable) exposure difference in the three dependent variables under analysis – CBE, SBC, and BUI. Preliminary assumption testing was conducted to check normality, linearity, univariate and multivariate outliers, homogeneity of variance-covariance matrices and multicollinearity with no serious violations noted. The difference between RTM’s types showed to be marginally significant on the combined dependent variables, F (12, 66.435) = 1.407, p=0.185, Wilks Lambda = 0.549, ^{Partial η2}= 0.181. When the results for the dependent variable where considered separately, considering a Bonferroni adjusted alpha level of 0.017, only BUI showed a marginal significant level, F (4, 27), p=0.099, η2 = 0.244.

Main effect and

interaction RTM Mean SD F df Sig Partial η2 Hypothesis

CBE Total (n=32) 3,68 1,660 0.789 4 0.543 0.105+ H4: Not supported

Planned (n=6) 3,72 1,604 4

Opportunistic (n=6) 3,65 1,636 4

Everyday (n=7) 4,59 1,689 4

Watchlist (n=5) 3,30 1,409 4

39

Brand-To-Brand (n=8) 3,13 1,899 4

BUI Total (n=32) 3,97 1,755 2.176 4 0.337 0.244+++ H5: Not supported

Planned (n=6) 3,50 1,557 4

Opportunistic (n=6) 3,50 1,930 4

Everyday (n=7) 5,54 1,122 4

Watchlist (n=5) 3,95 1,255 4

Brand-To-Brand (n=8) 3,31 1,985 4

SBC Total (n=32) 3,27 1,893 1.191 4 0.099 0.150++ H6:Partially supported

Planned (n=6) 3,12 1,909 4

Opportunistic (n=6) 2,79 2,128 4

Everyday (n=7) 4,59 1,946 4

Watchlist (n=5) 3,14 1,478 4

Brand-To-Brand (n=8) 2,68 1,777 4

Note: +small effect size; ++medium effect size; +++large effect size. CBE: Consumer-Brand Engagement, BUI: Brand Usage Intent, SBC: Self-Brand Connection, RTM: Real-Time Marketing, SD: Standard Deviation, η2: Partial Eta Squared.

Table 7 – Pre-test: MANOVA of RTM’s exposure on CBE, SBC and BUI

An important deep dive aspect is in the inspection of the mean scores. Results indicated that Everyday RTM scores higher than average in all dependent variables when compared with the average , paired with Planned RTM in CBE as visualized in Table 7. In practical terms these conclusions would assume no significant difference for the dependent variables except for BUI. However, this last finding might suggest otherwise.

Going back to the essence of Everyday RTM, it is based on unknown topics and micro events, which can indeed suggest a difference when considered - since being true in the moment might even feel like talking with a pear about what is actually going on. Besides, it is even more targeted. For this, the same analysis was run but separating the factor now into Everyday RTM and the other types of RTM. Based on the same principle of comparing effects on the three dependent variables, a MANOVA was carried out with the same preliminary assumptions verified.

As presented in Table 8, the results were fairly positive – even with the limitation of a small sample size. The difference between Everyday RTM and other RTM types showed to be significant on the combined dependent variables, F (3, 28) = 4.17, p=0.015, Wilks Lambda = 0.619, ^{Partial η2}= 0.309. When the results for the dependent variable where considered separately, using a Bonferroni adjusted alpha level of 0.017, BUI showed a significant level, F (1, 30) , p= 0.005, SBC showed a significant value using an alpha of 0.05, while CBE showed a marginal significance level. This is indeed the most important finding, since it validates that there isn’t a significant difference among all types (which would be really

40 hard to obtain) but Everyday RTM is the “antidote” we are looking for. Because this model shows a better fit, we will keep this for further analysis and adjust the framework accordingly.

Main effect and

interaction Everyday RTM Mean SD F df Sig Partial η2 Hypothesis

CBE Total (n=32) 3,68 1,660 2.814 1 0.104 0.086+ Partially

supported No exposure

(n=25) 3,43 1,595 1

Exposure

(n=7) 4,59 1,689 1

BUI Total (n=32) 3,97 1,755 8.984 1 0.005* 0.23+++ Supported

No exposure

(n=25) 3,53 1,657 1

Exposure

(n=7) 5,54 1,122 1

SBC Total (n=32) 3,27 1,893 4.902 1 0.035** 0.140++ Supported

No exposure

(n=25) 2,90 1,741 1

Exposure

(n=7) 4,59 1,946 1

Note: *significant p<0.017 **significant p<0.05. +small effect size; ++medium effect size; +++large effect size. CBE: Consumer-Brand Engagement, BUI: Consumer-Brand Usage Intent, SBC: Self-Consumer-Brand Connection, RTM: Real-Time Marketing, SD: Standard Deviation, η2: Partial Eta Squared.

Table 8 – Pre-test: Multiple analysis of variance of Everyday RTM exposure on CBE, SBC and BUI Furthermore, it is important to review the effect size scores – which are interpreted according to Maher et al. (2013) criteria for Eta Squared effect size measures. Results show that 8.6% of the variance in CBE is accounted by Everyday RTM Exposure, showing a medium effect size. On the other hand, 23%

of the variance in BUI is accounted for by Everyday RTM Exposure and 14% of the variance in SBC, demonstrating a large effect size for both of these dependent variables.

The not statistically significant result of Everyday RTM exposure on CBE might be an indication of the need for a mediation test. Technically, this was a fair assumption – since if there was a direct effect between these two variables, a hypothesis considering mediation wouldn’t be valid a priori.

5.5. BCE

E

RTM

CBE

Based on the previous results, a mediation analysis becomes even more important to access and validate the model. As mentioned, Waqas et al. (2021) stated that BCE was an antecedent of CBE,

41 meaning a good fit for mediation is expected. However, the interesting finding will be based on which factor(s) will impact this mediation.

Using Hayes’ Process macro (2017, model 4), with Everyday RTM exposure as the independent variable, CBE as the dependent variable, the four constructs of BCE were evaluated as possible moderators and the results and presented on Table 8.

BootLLCI BootULCI Total effect Indirect effect

% Indirect

effect Hypothesis

BCE_ID.Bound -0.7494 1.1062 Not supported

BCE_humor -0.2439 2.4873 Not supported

BCE_Aesth.Insp -0.6294 2.2740 Not supported

BCE_Discerning 0.1795* 1.7705* 1.1577 0.9862 85.19% Supported

Note: *significant p< 0.05, % Indirect effect = Indirect Effect/Total Effect, n = 32.

Table 9 – Pre-test: Evaluation of BCE factors as mediators

Contrary to what would be expected, the factor of BCE that mediates the relationship between Everyday RTM effect on CBE is Discerning and not Humor. We can assume a p value <0.05 given that the bootstrap for the upper and lower 95% confidence intervals are both with the same signal (in this case, suggesting a positive indirect effect) meaning we validate the hypothesis of an existing indirect effect of the Discerning component of BCE in the relationship of these two variables. Furthermore, calculating the percentage of indirect effect, we can conclude that the indirect effect of Discerning influenced 85.2% of the relationship between the dependent and independent variable - meaning that only 14.8% of the relation between RTM and CBE is explained without any mediation effect.

Again, “expecting the unexpected” is always a good assumption when it comes to testing new concepts and relations– and this conclusion does make sense when looked in a deeper perspective: we cannot make the mistake of interpreting the mediating assumption with the question “what makes participants engage the most?” since that is not the question (and it would make sense to be humor, since it would be an “in-the-moment” decision). The question is with what brands they engage the most, and it is a fair assumption to think users would interact more with brand that creates content around topics that are related to what they know and use their intellect to understand.

Concluding, we validate the BCE Discerning factor as mediator between Everyday RTM exposure and CBE, meaning that, the more the brand’s Everyday RTM content pushes for the reliability with the user’s interest and memory retrieval, the more it will have a positive perception and impact on CBE.

Therefore, only hypothesis H1e is accepted.

42

5.6. M

CBE

SBC

BUI

The last hypothesis to be tested aim to validate if there is any positive moderation between those who engage more with the brand after being exposed to Everyday RTM on SBC and BUI. Even though the impact of Everyday RTM on SBC and BUI is strong and statistically significant by itself, studying this effect might be relevant given Hollebeek et al. previous study accessing CBE as an antecedent of both constructs (2014).

Two separate moderation analysis were performed, using Hayes’ Process macro (2017, model 1), with Everyday RTM exposure as independent variable, SBC and BUI as the dependent variables, and CBE as the moderator. As shown in Table 10, the moderating effect on BUI revealed a marginally significant interaction (b = -0.36, SE = 0.21, t = -1.75 p = 0.09). The same analysis was performed with SBC as the dependent variable, revealing no statistically significant interaction (b = 0.09, SE = .13, t = 0.11 p = .91).

Dependent Variable b SE t Sig Covariate Assessment

BUI -0.3604 0.2058 -1.7510 0.0909 Partially supported

SBC 0.0217 0.1888 0.1149 0.9093 Not supported

Note: SE: Standard Error, BUI: Brand Usage Intent, SBC: Self-Brand Connection, n=32.

Table 10 – Pre-test: Moderation effect of CBE on BUI and SBC

This finding suggests that just because users engage with the brand when exposed to this RTM type, doesn’t mean they will instantly feel a connection towards the brand. However, it might be relevant to strength the effect on choosing this brand over others when considering or performing a purchase decision – a finding that can be better accessed when more responses are collected.

5.7. O

F

Given the previously accessed importance of OE and Generation (Millennials and Gen Z) in CBE, both these variables were evaluated as possible covariates. These are not relevant to the study itself but might have relevant implications that can be controlled. The assessment was made to explain if they would be covariates within the mediation process of BCE Discerning factor between Everyday RTM exposure and CBE. Therefore, using once again Hayes’ Process macro (2017, model 4), we aim to reject the null hypothesis which indicate that these wouldn’t be good covariate factors.

43 Findings suggest that OE is not a suitable covariate, but Generation presents a marginally significant p-value as observed in Table 11 - which means that maybe with more data we might be able to conclude that is can be an important factor to control for when accessing the relationship hypothesized between the variables.

Construct b SE t Sig Conclusion

Generation -2.7833 1.594 -1.747 0.0984 Partially Significant

OE -0.0619 0.3592 -0.1723 0.8645 Rejected

Note: SE: Standard Error, OE: Openness to experience, n=32.

Table 11 – Pre-test: Accessing Covariate Effect of OE and Generation in the mediating effect of BCE_Discerning in CBE

44